A repair request record is an unstructured short text written by customers to describe quality problems of building products. The extraction of entities and the construction of a hierarchical structure can provide an efficient source of knowledge for business decision and quality supervision. Firstly, based on the syntactic dependency and the lexical semantic information of repair request records, an approach combined the double propagation algorithm and the defect entity domain relevance is applied on repair request records to achieve entity extraction. Then, a graph-based hierarchy construction method is used to organize the extracted entities into a hierarchical structure. We further introduce the defect entity peer confidence to handle the task of graph pruning. Experiment results show that our methods can construct the hierarchy effectively and achieve better precision and recall compared to existing methods.